Simple Non‐Destructive and 3D Multi‐Layer Visual Hull Reconstruction with an Ultrabroadband Carbon Nanotubes Photo‐Imager

While millimeter‐wave (MMW)–infrared (IR) broad photo‐imaging exhibits significant potential as an inspection technique, photo‐thermoelectric (PTE) detector operations employing carbon nanotubes (CNTs) are highly suitable because of their efficient absorption characteristics in the above bands. Such an imaging device design and the associated material coordination allow the aggregation of multiple wavelength‐specific optical information for the constituent identification of target objects in a non‐destructive manner with high sensitivity. On the other side, computer vision techniques facilitate 3D structure reconstruction as approaches from inspection methodologies, in addition to the aforementioned imaging device design. Although non‐destructive inspections require collectively satisfying both constituent identification and structural reconstruction, investigations combining MMW–IR broad photo imaging and computer vision monitoring are still insufficient. This work demonstrates computer vision‐driven simple 3D composite multi‐layer reconstructions via non‐destructive broadband multiple‐wavelength monitoring with the CNT film PTE imager. Visual hull evaluation, which is one of the representative computer vision techniques, converts 2D transmissive PTE images with different viewpoints into 3D reconstruction models. The subsequential graphical superposition of these wavelength‐specific models results in non‐destructive entire reconstructions of 3D composite multi‐layer target objects. Therefore, the present computer vision‐driven PTE broadband 3D reconstruction bridges the gap between methodologies and device design strategies toward non‐destructive inspection applications.


Introduction
Photo monitoring techniques play an indispensable role in global mass production and distribution.Regarding the daily use of such industrial components, non-destructive inspections ensure fundamental safety.Among the representative non-destructive inspection techniques, including magnetic particle testing, [1,2] eddy current testing, [3,4] and penetrant testing, [5,6] photo monitoring techniques advantageously acquire optical information from large areas in a non-contact manner.Such large area acquisition of optical properties facilitates the conversion of inspection information to image data.Photo-imaging measurements enhance inspection efficiency over other single-point testing approaches, significantly contributing to non-destructive industrial inspections.Moreover, the non-contact operating configurations of photo-imaging measurements maintain the inherent behavior of inspecting targets and avoid unnecessary deterioration of objects due to the adhesion of reagents.Furthermore, the optical properties of the inspected targets and their constituent materials vary depending on the imaging wavelength.This implies that such nature of photo-imaging measurements allows the identification of the material composition of the inspection target via multiplewavelength monitoring. [7,8]Particularly, photo-imaging measurements from millimeter-wave (MMW) to terahertz (THz) and infrared (IR, far-IR: FIR, mid-IR: MIR, and near-IR: NIR) bands also exhibit transparency against non-metallic materials. [9,10]his feature indicates that photo-imaging measurements can play a leading role in non-destructive inspection techniques by obtaining inner images and identifying the concealed material composition of opaque target objects collectively in a non-contact manner.
These situations have accelerated the development of devices and systems for MMW-IR imaging measurements. [11,12]ocusing on studies regarding the development of image sensors, device operations under the photo-thermoelectric (PTE) effect, which is a synergetic phenomenon of photo-absorptioninduced heating and thermoelectric (TE) conversion, [13,14] potentially contribute to realizing non-destructive inspections in the MMW-IR bands.The PTE imager can be applied to uncooled broadband (MMW-IR) photo-detection. [15]Such broadband operations lead to the aggregation of optical information from multiple-wavelengths through compact singledevice configurations without switching among different image sensors.Moreover, uncooled device operations have resulted in the development of simple experimental setups that do not require bulky cooling equipment, facilitating onsite non-destructive MMW-IR imaging inspections.Therefore, appropriate device material selection for image sensors is indispensable for maximizing the potential advantages of PTE measurements for non-destructive inspection applications.While diverse materials are suitable for use as PTE imagers, including graphene, [16] black phosphorous, [17] transition metal dichalcogenide, [18] and Poly(3,4-ethylenedioxythiophene)poly(styrenesulfonate) (PEDOT:PSS), [19] carbon nanotube (CNT) films collectively satisfy the ultrabroadband (MMW-visible light (Vis)) highly efficient (over 90%) photo-absorption [20,21] over the above materials and the comparable Seebeck coefficient [22][23][24] with them.Indeed, the CNT film PTE imager exhibits comparable photo-detection sensitivities in broad MMW-IR regions, even with narrowband sensors in respective regions, [25,26] and demonstrates non-destructive material identification for inspecting target objects via multiple-wavelength monitoring. [27,28]espite the above, studies on demonstrating non-destructive 3D reconstructions of the inner structures of opaque inspection target objects via broad MMW-IR monitoring are still insufficient.Considering that industrial components typically exhibit composite multi-layer 3D structures, the fundamental shape or positioning reconstructions of the respective constituent elements are essential for non-destructive inspections, together with identifying the material composition.This situation is also a crucial bottleneck in developing the aforementioned CNT film PTE imager-based non-destructive inspection techniques.Employing multiple pixel-integrated CNT film imager sheets and their inherent mechanical softness provides omni-directional viewing angles without blind spots for monitoring 3D objects by freely adjusting the shape and size of the devices along with the target structures.However, such conventional omni-directional PTE imaging results in extracting different non-destructive inner hi-erarchical multi-view 2D images and performing their face-plane superposition from 3D objects at most. [29]In addition, information regarding the size or positioning of the respective inner elements of the composite multi-layer 3D target objects is crucially missing.Over time, progress in computer vision studies has facilitated fundamental 3D reconstructions with narrowband monitoring configurations. [30,31]Computer vision measurement is a representative 3D reconstruction approach.Computer vision-driven measurements typically handle fundamental photo signals (transmission, reflection, and scattering) or their associated changes and involve them under related conditions, such as intensity, angle, position, time delay, and phase shift. [32]In this mechanism, computer vision techniques provide various 3D reconstructed models, including outer-shape restorations, [33] surface observations, [34] and narrowband THz computer tomography. [35]Nonetheless, efforts to demonstrate the associated computer vision-driven non-destructive broadband (MMW-IR) multiple-wavelength imaging and the subsequent inner structure reconstructions of composite multi-layer 3D objects are insufficient, yet.In other words, demonstrating the above concept is indispensable for developing CNT film imagers and PTE monitoring as practical industrial non-destructive inspection techniques.
To this end, this work combines CNT film PTE imager-based MMW-IR visualization methods and visual hull-driven simple 3D reconstruction techniques for non-destructive monitoring of an opaque composite multi-layer target object (Figure 1a).Visual hull modeling is a widespread 3D reconstruction approach that links transmission intensities or their distribution images with respective angle dependences. [36]First, wavelength-specific hierarchical 2D images are selectively extracted from the composite multi-layer object via non-destructive ultrabroadband MMW-Vis monitoring (Figure 1b), and the respective 3D reconstructed models are superimposed into the composite multi-layer restoration (Figure 1c).This work reconstructs the inner structure of a multi-layer object comprising glass, semiconductors, plastic, and metal, which are the major constituent materials in today's industrial components.This point emphasizes that the present findings and techniques potentially enrich non-destructive inspection fields by demonstrating the concept of computer visiondriven PTE ultrabroadband 3D composite multi-layer reconstruction.

Results and Discussion
Figure 2a illustrates the experimental setup for the transmissive xy-scan PTE imaging measurements."Experimental Setup for Imaging Measurements" and "Photo Sources" in the Experimental Section provide detailed information for each equipment.The target object moves in the perpendicular plane direction (linear and rotating) to the optical path between the photo sources and the pn-junction type CNT film PTE imager (Figure 2b).
Under the imager operating mechanism of the PTE effect, the pn-junction in the CNT film channel serves as the photodetection interface and provides direct-current voltage (DCV) signals in response to external photo-irradiation onto it. [37]In Figure 2b, ∆V, S, and ∆T correspond to the PTE DCV response of the device against external irradiation, the Seebeck coefficient of the p-type (55 μV K −1 ) and n-type (−36 μV K −1 ) CNT films, and photo-induced temperature gradient across the channel, [38] respectively.The imager fabrication is available in versatile uncooled and air-exposed conditions by employing CNT solutions (Figure S1a, Supporting Information), and the device comprises a shape-and size-controllable channel (Figure S1b, Supporting Information).The device fabrication then moves to electrode wiring, chemical carrier dopant application (Figure S2a, Supporting Information), and mounting on a datalogger module (Figure S2b, Supporting Information)."Device Fabrication" in the Experimental Section further describes the detailed process.
Figure 2c depicts a schematic model of PTE conversion with the CNT film imager.As this device sets the pn-junction in the center of the CNT film channel as the photo-detection interface, the temperature becomes the highest there throughout the entire structure.This situation locally induces thermal carrier excitation around the photo-detection interface, and holes and electrons diffuse in opposite directions from the pn-junction.
While PTE detectors, inspired by typical TE modules, [39] often employ heterogeneous material junctions as the photo-detection interface, [13] the device comprising channels with the same polarity suppresses the total response where opposing electric fields cancel each other (Figure S3a, Supporting Information).In contrast, employing the pn-junction maximizes the total PTE response, which is greater than that with homogeneous material channels (Figure S3b, Supporting Information).Figure S4a,b Supporting Information also demonstrate the significance of the pn-junction for the CNT film PTE imager by clarifying response intensity distributions across the device structure.
The employed CNT film is a single-walled semiconductingmetallic-mixed type with randomly aligned microscopic network structures (Figure S5a, Supporting Information) and provides PTE responses under zero-voltage-bias operations (Figure S5b-d, Supporting Information).To form the photo-detection interface with the pn-junction, this work partially applies air-stable n-type chemical carrier dopant (see "Device Fabrication" in the Experimental Section) onto the p-type pristine CNT film (Figure S5e, Supporting Information).The above discussions represent the fundamental device operating mechanism utilized for the visual hull reconstruction in this work.Regarding the fundamental performance, the device first functions in a tens of milliseconds order (Figure S5f, Supporting Information), which is comparable with that of typical thermal type (uncooled broadband) photodetectors. [40]Meanwhile, photo-detection with the CNT film PTE imager is available in ultrabroadband regions over representative uncooled devices, with comparable sensitivities to them [41][42][43][44][45][46][47] (Figure S6, Supporting Information).Here, the authors' previous material and device design on the CNT film PTE imager [29] facilitates such advantageous performances.The previous work clarified that employing semiconducting-metallic-mixed compositions for the CNT film material type maximizes the PTE sensitivity among potential electronic states with optimal doping conditions (Figure S6, Supporting Information and see "Device Fabrication" in the Experimental Section).The CNT film PTE imager employed in this work consists of the same material and design conditions as the above, and this situation facilitates demonstrations of the presenting ultrabroadband 3D visual hull reconstruction technique.
For the visual hull reconstruction, the experimental system only requires utilizing the CNT film PTE imager for multiplewavelength measurements, with the device's inherent nature of ultrabroadband (MMW-Vis) highly efficient photo-absorption (Figure 2d).Furthermore, the CNT film PTE imager comprises a thin-sheet configuration with a thickness of tens of micrometers and functions even during mechanical deformations. [15]uch features advantageously allow freely-attachable device operations and ease the modeling of equipment setups depending on the experimental environment (Figure S7, Supporting Information).Typical approaches to broadband photo-sensing measurements involve the use of spectrometers.However, spectroscopic measurements often require setting targets within the equipment and regulating the samples' shape, structure, and size.However, the combined use of the patch-style CNT film PTE imager and multiple compact single-wavelength photo sources leads to a flexible experimental system design for diverse objects.In this work, spectroscopy measurements contributed to ultrabroadband multiple-wavelength PTE visual hull reconstructions by providing the fundamental optical properties of target objects and facilitating the proper choice of single-wavelength photo sources.
Changes in the output power of the external photo sources modulate the PTE response of the device (Figure 2e).Therefore, against the type of imaging objects, the device exhibits different PTE response intensities depending on the transmittances of the target materials and associated changes in the output power of the transmitted irradiation.In the transmissive imaging measurements, the changes in the PTE responses of the device correspond to the monochrome color scale.Figure 2f demonstrates transmissive xy-scan PTE imaging of a cuboid under THz irradiation.Here, the target is opaque to THz irradiation.This situation induces local reductions in transmission signals and the subsequent PTE response of the device, corresponding to the object' shape.The higher and lower intensities of the device responses correspond to the white and black color gradations in the ob-tained PTE image.By incorporating these conditions, the transmissive xy-scan measurement of the target cuboid results in the visualization of its silhouette (28-mm-length in the y-axis direction and 5-mm-width in the x-axis direction) in the PTE image.Figure S8 (Supporting Information) introduces the experimental setup for 2D-scanning and -rotation measurement, including a flowchart of the xy imaging.
Figure 3a illustrates the experimental procedure for a simple 3D visual hull reconstruction.This work considers cuboid structures as the reconstruction target, so the visual hull measurement employs two transmissive PTE silhouette images with a viewing angle difference of 90°.The covering length of the xy-scan imaging measurements was 35 mm.Therefore, the reference voxel was a cube with 35-mm-length on the side.The next step was to hollow the reference voxel from two viewing angles using the preobtained PTE silhouette images.In the hollowing-out process, the area where the respective silhouettes overlap within the reference voxel corresponded to the reconstructed structure (see Supporting Information for the 3D visual hull reconstruction code and "3D Visual Hull Reconstruction" in the Experimental Section for a detailed explanation).The hollowing-out process converted the PTE silhouette images from a monochrome color scale to a black-and-white color scale to extract the dominant opaque regions.Based on these conditions, Figure 3b shows a simple transmissive visual hull reconstruction of an actual 3D-structured target.The reconstructed 3D model captured the structural feature of the original target appropriately.Here, Table S1 (Supporting Information) is a dataset example obtained by the presenting xyscanning, and Figure 3c briefly describes the subsequential data processing on 2D silhouette imaging and 3D body reconstruction.Regarding the fundamental performances of the presenting device and system in visual hull-based 3D imaging, Figures S9 and S10 (Supporting Information) evaluate reconstruction accuracy and resolution.The presenting device and system exhibit reconstruction accuracy within 1.9% error, and the minimum detectable size resolution is 0.2 mm.
Following such preparations, Figure 3d-e introduce the final structure of the composite multi-layered 3D object for ultrabroadband multiple-wavelength non-destructive visual hull reconstructions.The target object comprised four different materials.First, the inner structure of the target object was invisible owing to an opaque outer coating.Behind the outer coating, midlayers consisting of a glass plate and a semiconductor board were present.The inner structure of the target object constituted a plastic housing.The target object was equipped with a metallic bar inside the plastic housing: the center position of its entire structure (see Figure S11, Supporting Information, and "Modeling of a Multi-Layer Object" in the Experimental Section for the detailed modeling conditions).
Figure 4a,b shows the evaluations of the optical characteristics of the material elements of the object, as the first preparation toward demonstrating the ultrabroadband multiple-wavelength simple non-destructive 3D visual hull reconstruction of a composite multi-layer target.Spectroscopy measurements were performed to understand the fundamental transparency of each material element in ultrabroadband regions (MMW-Vis), and "Spectroscopy Measurements" in the Experimental Section explains the detailed experimental conditions.This evaluation played an essential role in determining the suitability of the above Figure 4c-f compare the PTE silhouette images of the composite multi-layer cuboid object (Figure 3d) in respective Vis, IR, THz, and MMW bands from two different viewpoints with an angle difference of 90°(Figure 3e).In the Vis band, the obtained  PTE images (Figure 4c) captured the silhouette of the opaque outer coating at both viewing angles.The partial white slits in these PTE silhouette images correspond to tiny spatial gaps between the walls aligned in the plane-face and perpendicular directions.In the IR band, the obtained PTE images (Figure 4d) non-destructively visualized the inside of the opaque outer walls by transmission.From the "View 1" direction, the image captured the silhouette of the mid-layer semiconductor board, which did not exhibit transparency against the irradiated IR band.On the other hand, from the "View 2" direction, the image captured the silhouette of the inner plastic housing by transmitting the mid-layer glass plate.The black slit on the right-hand side of the plastic silhouette corresponded to the mid-layer silicon board aligned perpendicularly.In the THz band, the obtained PTE images (Figure 4e) also non-destructively visualized the inside of the opaque outer walls by transmitting them.From the "View 1" direction, the image captured the silhouette of the inner plastic housing by transmitting the mid-layer silicon board.The black slit on the left-hand side of the plastic silhouette corresponds to the mid-layer glass plate aligned perpendicularly.Meanwhile, from the "View 2" direction, the image captured the silhouette of the mid-layer glass plate, which did not exhibit transparency against the irradiated THz band.Finally, in the MMW band, the obtained PTE images (Figure 4f) further nondestructively visualized the inside of the plastic housing by collectively transmitting it together with the opaque outer coating and mid-layers.From both viewing angles, the images captured the silhouette of the metallic bar located at the center of the entire structure of the composite multi-layer target object.Based on these results, ultrabroadband multiple-wavelength transmissive imaging measurements of the composite multi-layer object allow for the non-destructive extraction of each constituent structure due to the differences in the optical properties of the respective materials.
Figure 5a-e finally perform the non-destructive 3D reconstruction of the composite multi-layer object by incorporating the simple visual hull approach and the obtained black and white color scale broadband and multiple-wavelength 2D PTE images.First, referring to the Vis images captured the opaque outer body in the original voxel (Figure 5a).Second, combining the IR and THz images visualizes the middle layer region covered with the slide glass and semiconductor board (Figure 5b) in the opaque outer coating (Figure 5a), and the inner plastic housing (Figure 5c) in the middle layer region.As respective IR and THz transparency against the target from "View 1" and "View 2" are complementary, the alternative utilization of the silhouette images in these bands selectively extracts the middle layer and inner bodies.Employing the remaining MMW images, the metallic bar was reconstructed at the deepest center position of the entire target structure (Figure 5d).Therefore, the superposition of these respectively-reconstructed models resulted in a comprehensive understanding of the structure of the tested composite multilayer object (Figure 5e) in a non-destructive monitoring manner.In the obtained 3D reconstruction model, the size and positioning of the respective constituent elements of the target were consistent with the designed structure, as shown in Figure 3d,e.This work defines the color and dot size of each 3D reconstructed model using the MeshLab software.
As the first step for exhibiting the significance of the computer vision-driven ultrabroadband multi-wavelength monitoring, this work basically demonstrated the 3D visual hull reconstruction with the single-pixel CNT film PTE imager.To shorten the reconstruction time, which potentially enhances the operating efficiency in future inspection applications, the use of multiple pixels-integrated imager devices is effective for simplifying spatial scanning operations.To this end, Figure S12 (Supporting Information) demonstrates the 3D visual hull reconstruction with a 40 pixels (0.65 mm pitch)-integrated CNT film PTE scanner.
In this configuration, the pixel arraying direction corresponds to the horizontal axis of PTE images.Therefore, the silhouette image acquisition only requires one-axis scanning of the target to be reconstructed, which is faster than the xy-scanning with the single-pixel CNT film PTE imager (Videos S1 and S2, Supporting Information).For such 3D visual hull reconstructions with the CNT film PTE scanner, Figure S13 (Supporting Information) evaluates the fundamental performances.The presenting system with the CNT film PTE scanner exhibits reconstruction accuracy within 2% error with the minimum detectable size resolution of 0.65 mm.These performances are comparable to those of the aforementioned 3D visual hull reconstruction system employing the single-pixel CNT film PTE imager.Figures S14 and S15 and Videos S3 and S4 (Supporting Information) further demonstrate that the measurement time of the CNT film PTE scanner-based 3D visual reconstruction is 150 s at the fastest.
Based on the above discussions, Figure S16 (Supporting Information) summarizes the presence of the presenting device and system among representative visual hull reconstruction techniques regarding operating wavelength bands and target materials or structures.While the representative visual hull techniques handle the outer shape of 3D objects under Vis illumination, this work extends their typical operating bands to permeable MMW-IR irradiation.Such an advantageous presence enabled the nondestructive structure reconstruction of a 3D composite object per each inner hierarchical layer.From a broader viewpoint, Table S2 (Supporting Information) compares detailed system performances and operating conditions among representative nondestructive structure reconstruction techniques (e.g., light detection and ranging (LiDAR), computed tomography (CT), photoacoustic, and so on).Among these, the presenting approach with advantageous non-invasive ultrabroad operations collectively satisfies inner material identifications and fundamental hierarchical 3D structure reconstructions with relatively reasonable system size and cost in ubiquitous conditions under comparable accuracy to that of others.

Conclusion
In conclusion, this work first demonstrates the synergetic effect of 3D computer vision techniques and MMW-IR ultrabroadband multiple-wavelength photo-monitoring with compact experimental setups employing the multi-functional CNT film PTE imager.Since industrial components often comprise multilayer structures with composite elements (glass, semiconductor, plastic, metal, etc.), as mentioned earlier, the subsequent nondestructive 3D reconstruction of the simple model consisting of the above materials plays a vital role.In other words, the proposed technique can potentially contribute to non-destructive industrial inspections.Furthermore, the successful combination of 3D computer vision techniques and CNT film PTE imager-based ultrabroadband multiple-wavelength photo-monitoring would lead to the development of other types of non-destructive reconstruction approaches (e.g., PTE CT).In addition to positioning and outline extraction in this work, increasing the viewing angles from the current two faces, such as PTE CT monitoring, facilitates the realization of detailed shape reconstructions from 3D composite multi-layer structures.

Figure 1 .
Figure 1.Conceptual diagram of this work.a,b) Schematic flow of a) ultrabroadband visual hull reconstruction and b) wavelength-specific extraction of a 3D multi-layer object.c) Simple comparison of the 3D multi-layer object before and after the ultrabroadband visual hull reconstruction.

Figure 2 .
Figure 2. Fundamental properties of the CNT film PTE imager.a) Experimental setup for PTE transmissive xy-scanning measurements of a 3D multilayer object with the CNT film imager toward the ultrabroadband visual hull reconstruction.b) Operating mechanism of the CNT film ultrabroadband PTE imager.c) Schematic model of the PTE effect.d) Optical properties of the CNT film.e) Ultrabroadband uncooled photo-detection with the device under the PTE effect.f) Example of the PTE transmissive xy-scan imaging measurement of a cuboid with the device under external THz ( = 10.3 μm) irradiation.

Figure 3 .
Figure 3. Fundamental mechanism of visual hull reconstruction.a) Experimental flow for the simple visual hull reconstruction.b) Example of the simple 3D visual hull reconstruction of a cuboid under external THz ( = 10.3 μm) irradiation.c) Simple flowchart of the presenting 3D visual hull reconstruction.d) Photographs of the composite multi-layer target object.e) Schematic structure of the composite multi-layer object.

Figure 4 .
Figure 4. Non-destructive silhouette image acquisition of the composite multi-layer object (Figure 3d,e) via xy-scanning with the CNT film PTE sensor under external ultrabroadband irradiation.a) Optical characteristics of the respective material elements of the composite multi-layer target object.b) Transmittance of the respective material elements of the composite multi-layer object under external MMW ( = 1.15 mm) irradiation.c-f) Transmissive xy-scan PTE silhouette images of the target from two different viewing angles (Figure 3d) under external Vis (c,  = 660 nm), IR (d,  = 976 nm), THz (e,  = 10.3 μm), and MMW (f) irradiation.